Nate Silver Predicts the Tournament
By combining two things I geek out about -- politics and sports statistics -- Nate Silver has become something of a legend in my own mind. Plus he's from Michigan. Briefly, he's a baseball statistician who rose to prominance by predicting the 2008 election better than any polls and has since become a sort of pop-statistician for the New York Times doing both election and sports predictions. He has come out with a bracket largely based on KenPom and similar computer rankings adjusted for injuries, locations and the like (as far as I can tell it hasn't been adjusted for current form, which is why Villanova isn't given a 0% chance of winning their first game). It picks Tennessee to win, so I'm angry at it right now, but it's worth a look and at least as legit as Joe Lunardi.
http://ncaabracket.nytimes.com/2011/bracket/men/fivethirtyeight-forecast
we probably overestimate "recent form" quite a bit. Villanova might be an extreme case because they've been so very bad (honestly not sure), but as a general matter, the last few games aren't a very good predictor of how a team will do in the tourney.
Somebody needs to code a program such that whenever somebody posts a prediction that completely counts us out, it instantly posts a reply containing the video with the Fab Five saying "We gonna shock the world!"
HEAVY chalk all the way on that one, but I guess that's what you get with the KenPom or other statistical models
I find it hard to believe that there will only be 2 first round upsets.
It is impossible with all of the upsets. No matter how good you are, statistics always favor the better teams, but when it is one and done this does not always hold true with who wins 1 singular game. It would be much more true and accurate if each round was a series, not an individual game.
I found it interesting that in his bracket, his Sweet 16 was all 4 - 1 seeds, 4 - 2 seeds, 4 - 3 seeds, and yes, 4 - 4 seeds. Has that ever happened?
March 15th, 2011 at 11:01 AM ^
describes their critique. They pointed out why such a statistical view fails to be successful in predicting.
March 15th, 2011 at 11:57 AM ^
And when is the last time all #1 seeds made the final four? Purdue is the only non 1 or 2 seed to make the Elite 8.
The great thing about Silver's predictions is that they are probablistic -he doesn't say "this is the team that will win," he says "there is an x% chance this tram will win'."
<br>
<br>The careless argument "yeah but statistics suck because on any given day yada yada..." is not an effective criticism of Silver's method - he is estimating the probability of "any given day" arriving on our doorstep. He is essentially estimating the chance his prediction is WRONG.
<br>
<br>We may not like the % chance of an M win he comes up with, but at least he gives each team a shot.
<br>
<br>
but they are better at predicting things like who will do well in league play, since the those are the consequence of many games played, and don't hinge on the outcomes of single games.
Looks like he has essentially no upsets, with the Sweet 16 consistenting of nothing but the top 4 seeds. He might as well have used the equally accurate statistical rule of "hey let's just pick the higher seed" since they were seeded higher for a reason.
Dude should stick with politics.
He isn't predicting the winners; he's predicting the ODDS.
Look at the Southeast region for example. He has Pitt 'winning' the region, but with only a 33% likelihood. In other words, while they might be the most likely team to advance out of that region, the odds are 2-1 against them winning it. Florida is 5-1, Wisconsin 10-1, BYU 10-1, and K State (my pick) is 16-1 (/changes bracket).
Compare that to the East region, where OSU is even money to advance, and no one else is better than 15%, or to the West region, where Duke is 40% to advance and SDSU is 30(!!!) to advance.
Of course the top seeds are the most likely to advance. And if they were to play this tournament 1000 times, the #1 seeds would probably advance out of each region more than any other team in that region. But there is value in knowing where the odds lie.
- The Southeast region could be a clusterfuck. It has the lowest win likelihood in the first round for any 4 seed (Wisconsin is 66% to beat Belmont), 5 seed (K State is 58% to beat Utah State), 6 seed (St Johns is a 2-1 dog to Gonzaga) and 7 seed (UCLA is a 2-1 dog to Sparty).
- SDSU is the non-#1 with the best chance to make the Final Four (30%)
- 3-seed Purdue is nearly twice as likely to make the Final Four than 2-seed Notre Dame (20%-11%)
March 15th, 2011 at 11:55 AM ^
Why is Gonzaga a favorite in all these statistical predictions? I know St. Johns lost one of their best players, but I still think they're good enough to win that game, and maybe even beat BYU.
March 15th, 2011 at 10:02 AM ^
Why does Villanova have a 0% chance of winning?
March 15th, 2011 at 10:03 AM ^
NATE SILVER LOVES CHALK
March 15th, 2011 at 10:13 AM ^
Interesting breakdown. Kind of predictable after the first two rounds because this sytem is somewhat self-fulfilling in that higher-regarded teams have a higher probability of winning, so unless you allow for some "randomness" you are going to have the highest-4 seeds generally advance if the bracket is laid out somewhat logically. In this case, the top-4 seeds in each region largely make sense (most people complain about the 8-12 seedings more) both numerically and logically, so a chalk final seems inevitable.
March 15th, 2011 at 10:31 AM ^
He also gives us decent odds - 43.1%. Consider that compared to other teams, Hampton at .8%, shows that he expects the game to be competitives. Additionally, this model updates with upsets and should be interesting to watch and see how the odds change.
March 15th, 2011 at 12:58 PM ^
Well, considering an 8 seed has a 46% chance of beating a 9 historically, 43% isn't all that great.